janus-llm 3.0.0__tar.gz → 3.1.0__tar.gz

{janus_llm-3.0.0 → janus_llm-3.1.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: janus-llm
-Version: 3.0.0
+Version: 3.1.0
 Summary: A transcoding library using LLMs.
 Home-page: https://github.com/janus-llm/janus-llm
 License: Apache 2.0

{janus_llm-3.0.0 → janus_llm-3.1.0}/janus/__init__.py

@@ -5,7 +5,7 @@ from langchain_core._api.deprecation import LangChainDeprecationWarning
 from .converter.translate import Translator
 from .metrics import *  # noqa: F403
 
-__version__ = "3.0.0"
+__version__ = "3.1.0"
 
 # Ignoring a deprecation warning from langchain_core that I can't seem to hunt down
 warnings.filterwarnings("ignore", category=LangChainDeprecationWarning)

{janus_llm-3.0.0 → janus_llm-3.1.0}/janus/converter/diagram.py

@@ -5,6 +5,7 @@ from janus.converter.converter import run_if_changed
 from janus.converter.document import Documenter
 from janus.language.block import TranslatedCodeBlock
 from janus.llm.models_info import MODEL_PROMPT_ENGINES
+from janus.parsers.uml import UMLSyntaxParser
 from janus.utils.logger import create_logger
 
 log = create_logger(__name__)
@@ -39,6 +40,7 @@ class DiagramGenerator(Documenter):
         self._diagram_type = diagram_type
         self._add_documentation = add_documentation
         self._documenter = None
+        self._diagram_parser = UMLSyntaxParser(language="plantuml")
         if add_documentation:
             self._diagram_prompt_template_name = "diagram_with_documentation"
         else:
@@ -83,7 +85,7 @@ class DiagramGenerator(Documenter):
 
         self._parser.set_reference(block.original)
 
-        query_and_parse = self.diagram_prompt | self._llm | self._parser
+        query_and_parse = self.diagram_prompt | self._llm | self._diagram_parser
 
         if self._add_documentation:
             block.text = query_and_parse.invoke(

{janus_llm-3.0.0 → janus_llm-3.1.0}/janus/converter/requirements.py

@@ -31,13 +31,14 @@ class RequirementsDocumenter(Documenter):
         output_list = list()
         # For each chunk of code, get generation metadata, the text of the code,
         # and the LLM generated requirements
-        for child in block.children:
-            code = child.original.text
-            requirements = self._parser.parse_combined_output(child.complete_text)
+        blocks = [block for block in block.children] if len(block.children) else [block]
+        for block in blocks:
+            code = block.original.text
+            requirements = self._parser.parse_combined_output(block.complete_text)
             metadata = dict(
-                retries=child.total_retries,
-                cost=child.total_cost,
-                processing_time=child.processing_time,
+                retries=block.total_retries,
+                cost=block.total_cost,
+                processing_time=block.processing_time,
             )
             # Put them all in a top level 'output' key
             output_list.append(

janus_llm-3.1.0/janus/parsers/uml.py

@@ -0,0 +1,51 @@
+import re
+import subprocess  # nosec
+from pathlib import Path
+from typing import List, Tuple
+
+from langchain_core.exceptions import OutputParserException
+
+from janus.utils.logger import create_logger
+
+from .code_parser import CodeParser
+
+log = create_logger(__name__)
+
+
+class UMLSyntaxParser(CodeParser):
+    def _get_uml_output(self, file: str) -> Tuple[str, str]:
+        # NOTE: running subprocess with shell=False, added nosec to label that we know
+        # risk exists
+        try:
+            plantuml_path = Path.home().expanduser() / ".janus/lib/plantuml.jar"
+            res = subprocess.run(
+                ["java", "-jar", plantuml_path, file],
+                stdout=subprocess.PIPE,
+                stderr=subprocess.PIPE,
+            )  # nosec
+            stdout = res.stdout.decode("utf-8")
+            stderr = res.stderr.decode("utf-8")
+        except FileNotFoundError:
+            log.warning("Plant UML executable not found, skipping syntax check")
+            stdout = ""
+            stderr = ""
+        return stdout, stderr
+
+    def _get_errs(self, s: str) -> List[str]:
+        return [x.group() for x in re.finditer(r"Error (.*)\n", s)]
+
+    def parse(self, text: str) -> str:
+        text = super().parse(text)
+        janus_path = Path.home().expanduser() / Path(".janus")
+        if not janus_path.exists():
+            janus_path.mkdir()
+        temp_file_path = janus_path / "tmp.txt"
+        with open(temp_file_path, "w") as f:
+            f.write(text)
+        uml_std_out, uml_std_err = self._get_uml_output(temp_file_path)
+        uml_errs = self._get_errs(uml_std_out) + self._get_errs(uml_std_err)
+        if len(uml_errs) > 0:
+            raise OutputParserException(
+                "Error: Received UML Errors:\n" + "\n".join(uml_errs)
+            )
+        return text

{janus_llm-3.0.0 → janus_llm-3.1.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "janus-llm"
-version = "3.0.0"
+version = "3.1.0"
 description = "A transcoding library using LLMs."
 authors = ["Michael Doyle <mdoyle@mitre.org>", "Chris Glasz <cglasz@mitre.org>",
            "Chris Tohline <ctohline@mitre.org>", "William Macke <wmacke@mitre.org>",